X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/43538243995267c417983360d226d6e8eb181139..5a662d84f00c0c2693c18d333bab9d0fdda7e28e:/services/datamanager/datamanager.go

diff --git a/services/datamanager/datamanager.go b/services/datamanager/datamanager.go
index bd68db112a..d3efe62173 100644
--- a/services/datamanager/datamanager.go
+++ b/services/datamanager/datamanager.go
@@ -10,29 +10,48 @@ import (
 	"git.curoverse.com/arvados.git/services/datamanager/collection"
 	"git.curoverse.com/arvados.git/services/datamanager/keep"
 	"git.curoverse.com/arvados.git/services/datamanager/loggerutil"
+	"git.curoverse.com/arvados.git/services/datamanager/summary"
 	"log"
 	"time"
 )
 
 var (
-	logEventType        string
+	logEventTypePrefix  string
 	logFrequencySeconds int
+	minutesBetweenRuns  int
 )
 
 func init() {
-	flag.StringVar(&logEventType,
-		"log-event-type",
-		"experimental-data-manager-report",
-		"event_type to use in our arvados log entries. Set to empty to turn off logging")
+	flag.StringVar(&logEventTypePrefix,
+		"log-event-type-prefix",
+		"experimental-data-manager",
+		"Prefix to use in the event_type of our arvados log entries. Set to empty to turn off logging")
 	flag.IntVar(&logFrequencySeconds,
 		"log-frequency-seconds",
 		20,
 		"How frequently we'll write log entries in seconds.")
+	flag.IntVar(&minutesBetweenRuns,
+		"minutes-between-runs",
+		0,
+		"How many minutes we wait betwen data manager runs. 0 means run once and exit.")
 }
 
 func main() {
 	flag.Parse()
+	if minutesBetweenRuns == 0 {
+		singlerun()
+	} else {
+		waitTime := time.Minute * time.Duration(minutesBetweenRuns)
+		for {
+			log.Println("Beginning Run")
+			singlerun()
+			log.Printf("Sleeping for %d minutes", minutesBetweenRuns)
+			time.Sleep(waitTime)
+		}
+	}
+}
 
+func singlerun() {
 	arv, err := arvadosclient.MakeArvadosClient()
 	if err != nil {
 		log.Fatalf("Error setting up arvados client %s", err.Error())
@@ -45,10 +64,11 @@ func main() {
 	}
 
 	var arvLogger *logger.Logger
-	if logEventType != "" {
-		arvLogger = logger.NewLogger(logger.LoggerParams{Client: arv,
-			EventType:     logEventType,
-			WriteInterval: time.Second * time.Duration(logFrequencySeconds)})
+	if logEventTypePrefix != "" {
+		arvLogger = logger.NewLogger(logger.LoggerParams{
+			Client:          arv,
+			EventTypePrefix: logEventTypePrefix,
+			WriteInterval:   time.Second * time.Duration(logFrequencySeconds)})
 	}
 
 	loggerutil.LogRunInfo(arvLogger)
@@ -56,28 +76,62 @@ func main() {
 		arvLogger.AddWriteHook(loggerutil.LogMemoryAlloc)
 	}
 
-	collectionChannel := make(chan collection.ReadCollections)
+	var (
+		readCollections collection.ReadCollections
+		keepServerInfo  keep.ReadServers
+	)
+
+	if !summary.MaybeReadData(arvLogger, &readCollections, &keepServerInfo) {
+		collectionChannel := make(chan collection.ReadCollections)
+
+		go func() {
+			collectionChannel <- collection.GetCollectionsAndSummarize(
+				collection.GetCollectionsParams{
+					Client:    arv,
+					Logger:    arvLogger,
+					BatchSize: 50})
+		}()
+
+		keepServerInfo = keep.GetKeepServersAndSummarize(
+			keep.GetKeepServersParams{
+				Client: arv,
+				Logger: arvLogger,
+				Limit:  1000})
+
+		readCollections = <-collectionChannel
+	}
+
+	summary.MaybeWriteData(arvLogger, readCollections, keepServerInfo)
 
-	go func() {
-		collectionChannel <- collection.GetCollectionsAndSummarize(
-			collection.GetCollectionsParams{
-				Client: arv, Logger: arvLogger, BatchSize: 50})
-	}()
+	buckets := summary.BucketReplication(readCollections, keepServerInfo)
+	bucketCounts := buckets.Counts()
 
-	keepServerInfo := keep.GetKeepServersAndSummarize(
-		keep.GetKeepServersParams{Client: arv, Logger: arvLogger, Limit: 1000})
+	replicationSummary := buckets.SummarizeBuckets(readCollections)
+	replicationCounts := replicationSummary.ComputeCounts()
 
-	readCollections := <-collectionChannel
+	log.Printf("Blocks In Collections: %d, "+
+		"\nBlocks In Keep: %d.",
+		len(readCollections.BlockToReplication),
+		len(keepServerInfo.BlockToServers))
+	log.Println(replicationCounts.PrettyPrint())
 
-	// TODO(misha): Use these together to verify replication.
-	_ = readCollections
-	_ = keepServerInfo
+	log.Printf("Blocks Histogram:")
+	for _, rlbss := range bucketCounts {
+		log.Printf("%+v: %10d",
+			rlbss.Levels,
+			rlbss.Count)
+	}
 
 	// Log that we're finished. We force the recording, since go will
-	// not wait for the timer before exiting.
+	// not wait for the write timer before exiting.
 	if arvLogger != nil {
 		arvLogger.FinalUpdate(func(p map[string]interface{}, e map[string]interface{}) {
-			p["run_info"].(map[string]interface{})["time_finished"] = time.Now()
+			summaryInfo := logger.GetOrCreateMap(p, "summary_info")
+			summaryInfo["block_replication_counts"] = bucketCounts
+			summaryInfo["replication_summary"] = replicationCounts
+			p["summary_info"] = summaryInfo
+
+			p["run_info"].(map[string]interface{})["finished_at"] = time.Now()
 		})
 	}
 }